Methods

Open Loop Lab uses human-AI collaboration as part of the research method, not as a hidden productivity layer. AI systems help generate possibilities, structure evidence, write and inspect code, stress-test arguments, and revise drafts. Human judgment governs what counts as evidence, what assumptions are acceptable, and what claims are warranted.

Research design and framing

Projects begin with a problem, not a favored method. The first task is to clarify what is actually being asked, what kind of evidence could answer it, and what existing literatures have already made the question more complicated.

This stage is deliberately iterative. Questions are reframed as the literature, data, and modeling constraints become clearer. The loop stays open because premature closure is one of the easiest ways to produce clean but weak research.

Evidence synthesis

Evidence synthesis is treated as infrastructure. Sources are collected, screened, coded, and organized into structured knowledge bases before analysis begins. The goal is not only to write a literature review, but to build a system that can be queried, updated, audited, and extended.

The Violence Project dashboard is the clearest current example: literature organized by ecological level and mechanism, built to support cumulative reasoning across levels of analysis.

Quantitative and computational modeling

The lab works in the GLM tradition, including mixed models, psychometrics, and structural equation modeling, and extends into computational modeling where the research question requires it.

Current modeling work includes path integral conflict models, Boltzmann-weighted basin probability estimation, counterfactual sensitivity analysis, and state vector construction from historical event data. R and Python are both used, with analysis pipelines built for reproducibility from the start.

Human-AI collaboration

AI systems participate in literature synthesis, code generation, methodological review, argument construction, and drafting. Their role is generative and analytic, but not authoritative.

A useful AI collaborator can widen the search space, reveal possible structures, accelerate implementation, and expose weaknesses in an argument. It can also introduce errors, false confidence, brittle abstractions, and fabricated evidence. The method depends on using the first set of capacities without hiding the second.

Where AI contributes substantively to a project, that contribution should be disclosed rather than obscured.

Tools and infrastructure

The working stack is designed for reproducibility and public communication:

Quarto for research documents and site generation
GitHub and GitHub Pages for version control and deployment
R and Python for analysis
Interactive HTML dashboards for research communication
VS Code as the primary development environment

The goal is to keep analysis, documentation, and publication close together, so the path from evidence to claim remains inspectable.

Transparency as method

Open Loop Lab treats transparency as a methodological requirement. Coding decisions, model assumptions, state vector dimensions, basin definitions, parameter choices, and AI contributions should be visible enough to inspect and contest.

A model that shows its assumptions can be argued with. That is the point.